Pitfalls of ascertainment biases in genome annotations—computing comparable protein domain distributions in eukarya

Most investigations into the large-scale patterns of protein evolution are based on gene annotations that have been compiled in reference databases. The use of these resources for quantitative comparisons, however, is complicated by sometimes vast differences in coverage. More importantly, however, we also observe substantial ascertainment biases that cannot be removed by simple normalization procedures. A striking example is provided by the correlations between protein domains. We observe that statistics derived from different computational gene annotation procedure show dramatic discrepancies, and even qualitative changes from negative to positive correlation, when compared to statistics obtained from annotation databases.